The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements a 'Decision feedback protocol' that executes a local Python script (utilities/skill-builder/scripts/record_skill_feedback.py) to persist interview outcomes. This command utilizes arguments such as notes and decisions derived from user interaction.
[PROMPT_INJECTION]: The skill is designed to process external documents or topics, which introduces a surface for indirect prompt injection.
Ingestion points: Document content or topics provided via the $ARGUMENTS variable.
Boundary markers: The skill lacks explicit delimiters or instructions to ignore embedded commands within the ingested content.
Capability inventory: The agent has the ability to append insights to local files and execute a local Python script via subprocess.
Sanitization: No specific sanitization or escaping of the processed document content is mentioned before it is summarized or used to generate interview questions.
[COMMAND_EXECUTION]: The skill reads local persona and configuration files from ~/.codex/USER_PROFILE.md and ~/.codex/AGENTS.md to align behavior with user preferences.

deep-interview